The Prisoner's Dilemma: When Two Rational People Both Lose

Two people are arrested for a crime they committed together and held in separate rooms. Each is offered the same deal: implicate your partner and you go free while they serve ten years. If both stay silent, the prosecutor can only prove a minor charge — one year each. If both implicate each other, both serve five.

Work through it from one prisoner's seat, assuming nothing about loyalty or ethics — only self-interest.

If my partner stays silent, I can serve one year by staying silent or walk free by talking. Talking is better.

If my partner talks, I can serve ten years by staying silent or five by talking. Talking is better.

Talking is better in both cases. It is what game theorists call a dominant strategy — the correct choice regardless of what the other person does. So a perfectly rational prisoner talks. And since the same reasoning applies to the partner, they talk too. Both serve five years.

Had they both stayed silent, both would have served one. The outcome available to them was four times better, and impeccable individual reasoning drove them away from it.

This is the most important twenty lines in social science, because the structure it describes is not a puzzle about prisoners. It is the shape of an enormous number of real situations in which everyone behaves sensibly and everyone ends up worse off.

What makes it a dilemma rather than a mistake

The essential feature — and the one most often missed — is that nobody in this scenario is being stupid, greedy, or short-sighted. There is no error to correct. Each player evaluates the situation accurately and chooses optimally given the incentives in front of them.

That is precisely what makes it structural. If the bad outcome came from a mistake, the fix would be education or better reasoning. It doesn't, so it isn't. The individually rational choice and the collectively rational choice genuinely diverge, and no amount of clear thinking by an individual closes the gap. A player who unilaterally cooperates in a one-shot game with a stranger simply gets exploited — that is not a moral failure on their part, it is the payoff matrix working as designed.

The structure appears wherever three conditions hold: mutual cooperation beats mutual defection; defecting against a cooperator pays best of all; and being the lone cooperator pays worst. Once you learn to recognise that shape, it becomes visible constantly.

Price competition. Two firms both profit more if both hold prices high. Either profits most by cutting while the other holds. Both cut. Both earn less than if neither had.

Arms races, whether military, advertising, or agricultural. Every participant would prefer that everyone spend less. Each individually benefits from spending more than rivals. All spend heavily and the relative positions are unchanged — the resources are simply consumed.

Shared resources. Every fisher would prefer sustainable stocks. Each individually benefits from catching more. This is the tragedy of the commons, which is a prisoner's dilemma played by many participants at once.

Working hours in competitive professions. Everyone would prefer a norm of reasonable hours. Anyone gains an edge by working more. The norm collapses toward exhaustion, and nobody can unilaterally opt out without paying for it.

The escape: repetition and the shadow of the future

The dilemma's grip depends on a hidden assumption — that the game is played once, between parties who will not meet again.

Change that and the mathematics changes fundamentally. In a repeated game with an indefinite horizon, today's betrayal has a future cost: the other player can retaliate in every subsequent round. If the relationship is likely to continue and the future matters enough, cooperating becomes the individually optimal strategy. Not because anyone became generous — because the accounting changed.

Robert Axelrod's tournaments in the early 1980s tested this by inviting researchers to submit strategies to play repeated prisoner's dilemmas against each other. The winner was strikingly simple: cooperate on the first move, then do whatever the opponent did last time. The lesson generally drawn is that successful strategies tend to be nice (never defect first), retaliatory (punish defection immediately), forgiving (return to cooperation once the opponent does), and clear (simple enough that the opponent can learn to predict you).

That final property is underrated. A strategy so complex that opponents cannot infer the pattern fails to teach them that cooperation pays — and teaching them is the entire mechanism by which cooperation becomes stable.

Institutions as payoff engineering

Once you see that cooperation emerges from repetition rather than virtue, most of human institutional design becomes legible as deliberate payoff modification. Every one of these mechanisms performs the same structural job: adding a cost to defection that the raw matrix lacked.

Contracts convert a one-shot interaction into a repeated one with the legal system as enforcer. The cost of defection stops being reputational and becomes financial and certain.

Reputation systems — professional licensing, credit ratings, marketplace reviews — make single interactions behave like repeated ones by ensuring that a defection against one counterparty is visible to all future counterparties. This is why anonymity so reliably degrades cooperation: it severs defection from consequence.

Regulation simply removes the defecting option, which is why industries prone to destructive competition frequently lobby for rules they would individually violate. Everyone would prefer the cooperative equilibrium; nobody can reach it alone.

Small, stable communities sustain cooperation with remarkably little formal enforcement, because everyone expects to meet everyone repeatedly. The shadow of the future is long by default. The corresponding failure mode is that cooperation degrades sharply in large, anonymous, high-turnover environments — not because people there are worse, but because the structure supporting cooperation is absent.

The practical diagnostic: when you see people behaving badly in a system, check the payoff matrix before checking their character. If defection dominates, you will observe defection regardless of who occupies the roles, and replacing the people changes nothing. The productive intervention is almost never exhortation — it is finding a way to make the future matter, or to make defection cost something.

Two prisoners with excellent reasoning and no way to make a binding promise to each other will serve five years apiece, every single time. The failure is not in them. It is in a situation that gave them no mechanism to commit — and building those mechanisms is, arguably, most of what civilisation is.